AI Fighters Unleashed: Creating Pro-Level Real-Time Combatants Using Deep Reinforcement Learning
Dr Raghvendra Chinchansoor, Dr Praveen B M · INTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 2025
This study presents the design and development of AI-driven combat agents capable of performing at a professional level in real-time fighting environments. By applying Deep Reinforcement Learning (DRL), specifically Proximal Policy Optimization (PPO), our system enables agents to autonomously learn complex tactics such as attack combos, blocking, dodging, and counter-attacks. Unlike traditional rule-based opponents, our AI fighters adapt dynamically through self-play and interaction with multiple adversaries. The training leverages a simulation environment with real-time physics, custom reward shaping, and multi-agent learning frameworks. Experimental results demonstrate that the trained agents not only surpass baseline scripted bots but also show emergent strategic behavior. This work has significant implications for AI in competitive gaming, robotics, and adaptive simulation systems. Keywords Deep Reinforcement Learning, Fighting Game AI, PPO, Real-Time Combat, Self-Play, Multi-Agent Systems, Game AI, Simulation, Policy Network, Emergent Strategy.